Visualizing Trips and Travel Characteristics from GPS Data

Stopher P, Bullock P and Jiang Q 2003 ‘Visualising trips and travel characteristics from GPS data‘, Road & Transport Research, vol.12:2, pp. 3-14.

This paper, in the field of automatic mobility surveys, talks about the issue in presenting information contained in GPS records so that it is understandable both to the survey respondent and to the decision makers. It relates on how to convert the data from cars travels into discrete trips, and on their visualization.

An example of the type of data available from the GPS devices consist in:

* Latitude and longitude in degrees and decimal degrees, with hemispheric (E, W, N,S) designation
* Altitude in meters above sea level
* Heading in degrees from north
* Coordinated Universal Time (UTC Time, or Greenwich Mean Time)
* Coordinated Universal Date (UTC Date)
* Speed in km/h
* Horizontal dispersion of precision (HDOP)
* Satellites in view.

The most challenging part of the data manipulation is breaking the record into individual trips. The main difficulty is to detect short stops, such as may occur when a person fills up the car with petrol, stops to post a letter or similar activities (e.g. when the engine is left running):

Such locations are found by removing from the data any data points where the movement between successive data points is less than the accuracy rating of the GPS device. In our case, the GPS devices in use are rated to have an accuracy of within + or -20 metres. However, when stationary, the position rarely changes by more than a few metres, with a speed of 0.0 kph, or nearly so. Therefore, by removing those points where the speed is shown to be zero, and there is little change in position, we can detect when there is a stop lasting two minutes or longer, and define that as a probable trip end.

The authors acknowledge the problems that my arise in the track records such as signal loss and warm-up time:

The devices in use are rated to acquire signal in 15-45 seconds, and generally succeeded, when stationary, to acquire position within no more than 15-20 seconds. However, if the device is immediately in motion when it is turned on, such as if an in-vehicle device is used and a person gets in the car and drives off immediately, a much longer time may be required to acquire position. This results from the vehicle motion, which requires the device to take longer to fix its position, and may also be exacerbated if there are interruptions to the signal, resulting from tall buildings or heavy tree canopies, while the device is attempting to acquire position. In our experiments, we found that in-motion acquisition time depended on how long it had been since the device was last turned on. For elapsed stop times of less than an hour or two, signal acquisition was still relatively fast and generally about 30-60 seconds. However, if a longer time had elapsed since the last use, the position acquisition could become lengthy, and exceeded, in a few cases, 1 kilometre of travel distance and about 2 minutes or more of time.

To develop an automated procedure to analyse the data, a rules based algorithm is suggested as follow:

• The difference in successive latitude and longitude values is less than 0.000 051 degrees; and
• The heading is unchanged or is zero; and
• Speed is zero; and
• Elapsed time during which these conditions hold is equal to or greater than 120 seconds.

If there is a break in the record, meaning that the engine was turned off, of between 30 and 120 seconds, this is also defined as a potential trip end.

From the analysis of known trips recorded by GPS and the application of the rules discussed in this paper, the authors determine that their is about a five purcent error rate in both detecting false trip ends, and in failing to detect real trip ends.

Relation to my thesis: With “Learning Significant Locations and Predicting User Movement with GPS“, “Elimination of the Travel Diary: An Experiment to Derive Trip Purpose from GPS Travel Data” and “Exploring the Potentials of Automatically Collected GPS Data for Travel Behaviours Analysis” I explored issues in manipulating user generated location information in the transportation planning research. The main problems with location quality depends on the sensed-data (e.g. premature end of data-stream due to urban canyon), the GIS database and the data processing (e.g. coding of trips).